Demographics

By sample

Demographics for included participants, by handedness sample: pilot righties, new righties, and lefties/mixedies.

Sample N (keepers) Age (years) Education (years) Sex (M/F/O) EHI
Left/mixedies 312 29 (6.01) 14.38 (2.3) 180/126/6 -28.65 (66.92)
New righties 205 29.58 (6.03) 14.13 (2.5) 104/99/2 90.06 (18.72)
Pilot righties 104 29.48 (5.95) 14.46 (2.48) 52/51/1 89.54 (14.79)
Left: (EHI <= -40) | Mixed: (-40 < EHI < 40) | Right: (EHI >= 40)


By handedness group

Demographics for included participants, by handedness group (EHI bins).

Handedness N Age (years) Education (years) Sex (M/F/O) EHI
Left 171 29.14 (6.27) 14.25 (2.27) 90/79/2 -80.19 (19.96)
Mixed 78 29.08 (6.08) 14.59 (2.32) 49/27/2 -5.61 (26.7)
Right 372 29.37 (5.88) 14.28 (2.47) 197/170/5 88.68 (15.52)
Left: (EHI <= -40) | Mixed: (-40 < EHI < 40) | Right: (EHI >= 40)


We have recruited 336 “Right handed”, and 366 “Left handed” or “Ambidextrous” participants, using Prolific’s pre-screeners.

There are fewer EHI-confirmed left and mixed handers than I expected. This is, in part, because some of Prolific’s pre-screened “Left-handed” and “Ambidextrous” participants are fairly right handed, as measured by the EHI:

If we want to get balanced handedness groups (a more U-shaped EHI distribution) on Prolific going forward, we should recruit a larger proportion of “Left-handed” and “Ambidextrous” participants.

Field x Level

Pilot righties (n = 104/112 RH)

In our pilot sample of right handers, do we see the typical field x level interaction? That is, do participants show a relative bias for global shapes in the left visual field (LVF)?

Summary. We see the predicted effect, for both reaction time (28.35ms, 95%CI [13.54, 43.16], p < .001) and accuracy (OR = 1.6, 95%CI [1.22, 2.28])

Reaction time

Plots

Statistics

Reaction time is modeled as a linear effect of field and level, using data from every target-present trial with a “go” response:

lmer( rt ~ field + level + field:level + (1 | subject) )


Field by level interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value
5 171,124.179 171,161.383 −85,557.089 171,114.179 - - -
6 171,112.1 171,156.746 −85,550.05 171,100.1 14.079 1 .0002


Field by level interaction (RT)
Omnibus F-test
term df sumsq meansq statistic p.value
field 1 16,788.12 16,788.12 0.226 .63
level 1 1,663,708.431 1,663,708.431 22.38 <.0001
field:level 1 512,981.705 512,981.705 6.9 .009
Residuals 12,590 935,944,119.098 74,340.28 - -


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global 28.351 7.555 Inf 13.544 43.158 3.753 .0002
1 A positive number means global bias is stronger in LVF (as predicted for right handers)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


Global bias by field (RT)
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local - LVF Global 36.765 5.341 Inf 26.296 47.233 6.883 <.0001
RVF Local - RVF Global 8.413 5.347 Inf −2.067 18.893 1.573 .12
1 A positive number means global bias (faster RT for global)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


RT estimates by field and level (from model)
field level emmean SE df1 asymp.LCL2 asymp.UCL2
LVF Global 672.671 17.273 Inf 638.817 706.525
LVF Local 709.436 17.29 Inf 675.548 743.323
RVF Global 688.52 17.278 Inf 654.656 722.385
RVF Local 696.934 17.286 Inf 663.054 730.813
1 Z-approximation
2 Confidence level: 95%


RT estimates by field and level (descriptive)
field level median mean SE
LVF Global 629 671.09 4.725
LVF Local 669 706.834 4.945
RVF Global 649 685.85 4.883
RVF Local 658 696.062 4.887


Accuracy

Plots

Statistics

Accuracy is modeled as a binomial effect of field and level, using binary correct/incorrect data from every target-present trial:

glmer( correct ~ field + level + field:level + (1 | subject), family = "binomial" )


Field by level interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
4 5,192.89 5,222.875 −2,592.445 5,184.89 - - -
5 5,185.098 5,222.58 −2,587.549 5,175.098 9.792 1 0.002


Field by level interaction (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec odds.ratio1 SE df2 asymp.LCL3 asymp.UCL null z.ratio p.value4
RVF / LVF Local / Global 1.664 0.267 Inf 1.215 2.277 1 3.177 .001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF, as predicted for right handers.
2 I don't understand why df is 'Inf' here, but I think it is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


Global bias by field (Accuracy)
contrast odds.ratio1 SE df asymp.LCL2 asymp.UCL null z.ratio p.value3
LVF Global / LVF Local 2.287 0.268 Inf 1.817 2.878 1 7.054 <.0001
RVF Global / RVF Local 1.375 0.15 Inf 1.11 1.703 1 2.914 .004
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias (more correct responses for global).
2 Confidence level: 95%
3 Two-sided, uncorrected


Accuracy estimates by field and level (from model)
field level prob1 SE df asymp.LCL2 asymp.UCL
LVF Global 0.977 0.003 Inf 0.97 0.983
LVF Local 0.949 0.006 Inf 0.936 0.96
RVF Global 0.968 0.004 Inf 0.959 0.975
RVF Local 0.956 0.005 Inf 0.945 0.965
1 Back-transformed to probability (% correct) from logit scale
2 Confidence level: 95%


Accuracy estimates by field and level (descriptive)
field level mean_subject_percent_correct
LVF Global 96.605
LVF Local 92.819
RVF Global 95.282
RVF Local 93.72


New righties (n = 205/224)

Do we replicate the field x level interaction in our new righties?

Summary. We see the predicted effect, for both reaction time (27.27ms, 95%CI [17.17, 36.37], p < .001) and accuracy (OR = 2.0, 95%CI [1.6, 2.6])

Reaction time

Plots

Statistics

Reaction time is modeled as a linear effect of field and level, using data from every target-present trial with a “go” response:

lmer( rt ~ field + level + field:level + (1 | subject) )


Field by level interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value
5 338,385.02 338,425.663 −169,187.51 338,375.02 - - -
6 338,359.032 338,407.803 −169,173.516 338,347.032 27.988 1 <.0001


Field by level interaction (RT)
Omnibus F-test
term df sumsq meansq statistic p.value
field 1 60.732 60.732 0.001 .98
level 1 2,006,529.632 2,006,529.632 28.683 <.0001
field:level 1 1,169,823.852 1,169,823.852 16.722 <.0001
Residuals 25,043 1,751,919,309.782 69,956.447 - -


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global 27.267 5.153 Inf 17.167 37.366 5.292 <.0001
1 A positive number means global bias is stronger in LVF (as predicted for right handers)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


Global bias by field (RT)
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local - LVF Global 32.585 3.638 Inf 25.455 39.715 8.957 <.0001
RVF Local - RVF Global 5.319 3.651 Inf −1.837 12.474 1.457 .15
1 A positive number means global bias (faster RT for global)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


RT estimates by field and level (from model)
field level emmean SE df1 asymp.LCL2 asymp.UCL2
LVF Global 675.866 12.113 Inf 652.125 699.608
LVF Local 708.452 12.123 Inf 684.69 732.213
RVF Global 690.156 12.119 Inf 666.404 713.909
RVF Local 695.475 12.122 Inf 671.716 719.234
1 Z-approximation
2 Confidence level: 95%


RT estimates by field and level (descriptive)
field level median mean SE
LVF Global 635 674.672 3.266
LVF Local 663 706.194 3.403
RVF Global 646 687.972 3.379
RVF Local 657 692.155 3.324


Accuracy

Plots

Statistics

Accuracy is modeled as a binomial effect of field and level, using binary correct/incorrect data from every target-present trial:

glmer( correct ~ field + level + field:level + (1 | subject), family = "binomial" )


Field by level interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
4 8,994.277 9,026.977 −4,493.139 8,986.277 - - -
5 8,964.187 9,005.063 −4,477.094 8,954.187 32.09 1 0


Field by level interaction (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec odds.ratio1 SE df2 asymp.LCL3 asymp.UCL null z.ratio p.value4
RVF / LVF Local / Global 2.048 0.256 Inf 1.603 2.617 1 5.733 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF, as predicted for right handers.
2 I don't understand why df is 'Inf' here, but I think it is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


Global bias by field (Accuracy)
contrast odds.ratio1 SE df asymp.LCL2 asymp.UCL null z.ratio p.value3
LVF Global / LVF Local 2.571 0.244 Inf 2.134 3.097 1 9.945 <.0001
RVF Global / RVF Local 1.255 0.102 Inf 1.07 1.472 1 2.79 .005
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias (more correct responses for global).
2 Confidence level: 95%
3 Two-sided, uncorrected


Accuracy estimates by field and level (from model)
field level prob1 SE df asymp.LCL2 asymp.UCL
LVF Global 0.984 0.002 Inf 0.98 0.987
LVF Local 0.96 0.004 Inf 0.952 0.966
RVF Global 0.971 0.003 Inf 0.965 0.976
RVF Local 0.964 0.003 Inf 0.957 0.969
1 Back-transformed to probability (% correct) from logit scale
2 Confidence level: 95%


Accuracy estimates by field and level (descriptive)
field level mean_subject_percent_correct
LVF Global 97.53
LVF Local 94.101
RVF Global 95.595
RVF Local 94.588


All righties (n = 309/336)

In our pilot sample of right handers, do we see the typical field x level interaction? That is, do participants show a relative bias for global shapes in the left visual field (LVF)?

Summary. We see the predicted effect, for both reaction time (27.63ms, 95%CI [19.28, 35.98], p < .001) and accuracy (OR = 1.89, 95%CI [1.56, 2.30])

Reaction time

Plots

Statistics

Reaction time is modeled as a linear effect of field and level, using data from every target-present trial with a “go” response:

lmer( rt ~ field + level + field:level + (1 | subject) )


Field by level interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value
5 509,525.115 509,567.795 −254,757.558 509,515.115 - - -
6 509,485.071 509,536.286 −254,736.535 509,473.071 42.045 1 <.0001


Field by level interaction (RT)
Omnibus F-test
term df sumsq meansq statistic p.value
field 1 4,699.114 4,699.114 0.066 .8
level 1 3,616,074.884 3,616,074.884 50.633 <.0001
field:level 1 1,680,647.538 1,680,647.538 23.533 <.0001
Residuals 37,637 2,687,932,890.743 71,417.299 - -


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global 27.627 4.26 Inf 19.278 35.976 6.486 <.0001
1 A positive number means global bias is stronger in LVF (as predicted for right handers)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


Global bias by field (RT)
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local - LVF Global 33.981 3.009 Inf 28.084 39.878 11.294 <.0001
RVF Local - RVF Global 6.354 3.017 Inf 0.441 12.267 2.106 .04
1 A positive number means global bias (faster RT for global)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


RT estimates by field and level (from model)
field level emmean SE df1 asymp.LCL2 asymp.UCL2
LVF Global 674.796 9.903 Inf 655.387 694.206
LVF Local 708.777 9.912 Inf 689.351 728.204
RVF Global 689.606 9.907 Inf 670.189 709.024
RVF Local 695.96 9.91 Inf 676.537 715.384
1 Z-approximation
2 Confidence level: 95%


RT estimates by field and level (descriptive)
field level median mean SE
LVF Global 633 673.474 2.687
LVF Local 665.5 706.407 2.804
RVF Global 646 687.259 2.78
RVF Local 658 693.462 2.75


Accuracy

Plots

Statistics

Accuracy is modeled as a binomial effect of field and level, using binary correct/incorrect data from every target-present trial:

glmer( correct ~ field + level + field:level + (1 | subject), family = "binomial" )


Field by level interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
4 14,183.391 14,217.732 −7,087.695 14,175.391 - - -
5 14,144.539 14,187.466 −7,067.269 14,134.539 40.852 1 0


Field by level interaction (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec odds.ratio1 SE df2 asymp.LCL3 asymp.UCL null z.ratio p.value4
RVF / LVF Local / Global 1.893 0.186 Inf 1.56 2.296 1 6.477 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF, as predicted for right handers.
2 I don't understand why df is 'Inf' here, but I think it is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


Global bias by field (Accuracy)
contrast odds.ratio1 SE df asymp.LCL2 asymp.UCL null z.ratio p.value3
LVF Global / LVF Local 2.455 0.181 Inf 2.124 2.837 1 12.169 <.0001
RVF Global / RVF Local 1.297 0.085 Inf 1.141 1.474 1 3.984 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias (more correct responses for global).
2 Confidence level: 95%
3 Two-sided, uncorrected


Accuracy estimates by field and level (from model)
field level prob1 SE df asymp.LCL2 asymp.UCL
LVF Global 0.982 0.002 Inf 0.979 0.985
LVF Local 0.957 0.003 Inf 0.95 0.962
RVF Global 0.97 0.002 Inf 0.965 0.974
RVF Local 0.961 0.003 Inf 0.955 0.966
1 Back-transformed to probability (% correct) from logit scale
2 Confidence level: 95%


Accuracy estimates by field and level (descriptive)
field level mean_subject_percent_correct
LVF Global 97.219
LVF Local 93.669
RVF Global 95.489
RVF Local 94.296


Lefties/mixedies (n = 312/336)

In our pilot sample of right handers, do we see the typical field x level interaction? That is, do participants show a relative bias for global shapes in the left visual field (LVF)?

Summary. We see an effect in the same direction as right handers’, for both reaction time (19.85ms, 95%CI [11.49, 28.22], p < .001) and accuracy (OR = 1.94, 95%CI [1.61, 2.35])

Reaction time

Plots

Statistics

Reaction time is modeled as a linear effect of field and level, using data from every target-present trial with a “go” response:

lmer( rt ~ field + level + field:level + (1 | subject) )


Field by level interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value
5 512,128.622 512,171.324 −256,059.311 512,118.622 - - -
6 512,109.005 512,160.248 −256,048.503 512,097.005 21.617 1 <.0001


Field by level interaction (RT)
Omnibus F-test
term df sumsq meansq statistic p.value
field 1 1,086,635.892 1,086,635.892 15.996 <.0001
level 1 5,787,209 5,787,209 85.192 <.0001
field:level 1 1,123,678.327 1,123,678.327 16.541 <.0001
Residuals 37,809 2,568,431,037.977 67,931.737 - -


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global 19.854 4.27 Inf 11.486 28.223 4.65 <.0001
1 A positive number means global bias is stronger in LVF (as predicted for right handers)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


Global bias by field (RT)
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local - LVF Global 34.096 3.021 Inf 28.176 40.017 11.287 <.0001
RVF Local - RVF Global 14.242 3.022 Inf 8.319 20.165 4.713 <.0001
1 A positive number means global bias (faster RT for global)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


RT estimates by field and level (from model)
field level emmean SE df1 asymp.LCL2 asymp.UCL2
LVF Global 653.142 9.222 Inf 635.068 671.216
LVF Local 687.238 9.235 Inf 669.138 705.339
RVF Global 674.032 9.225 Inf 655.95 692.113
RVF Local 688.274 9.232 Inf 670.18 706.367
1 Z-approximation
2 Confidence level: 95%


RT estimates by field and level (descriptive)
field level median mean SE
LVF Global 603 651.037 2.579
LVF Local 647 686.683 2.744
RVF Global 627 672.286 2.71
RVF Local 644 686.122 2.693


Accuracy

Plots

Statistics

Accuracy is modeled as a binomial effect of field and level, using binary correct/incorrect data from every target-present trial:

glmer( correct ~ field + level + field:level + (1 | subject), family = "binomial" )


Field by level interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
4 14,943.972 14,978.352 −7,467.986 14,935.972 - - -
5 14,899.846 14,942.822 −7,444.923 14,889.846 46.125 1 0


Field by level interaction (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec odds.ratio1 SE df2 asymp.LCL3 asymp.UCL null z.ratio p.value4
RVF / LVF Local / Global 1.944 0.188 Inf 1.609 2.349 1 6.885 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF, as predicted for right handers.
2 I don't understand why df is 'Inf' here, but I think it is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


Global bias by field (Accuracy)
contrast odds.ratio1 SE df asymp.LCL2 asymp.UCL null z.ratio p.value3
LVF Global / LVF Local 3.077 0.221 Inf 2.673 3.542 1 15.641 <.0001
RVF Global / RVF Local 1.583 0.102 Inf 1.394 1.796 1 7.098 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias (more correct responses for global).
2 Confidence level: 95%
3 Two-sided, uncorrected


Accuracy estimates by field and level (from model)
field level prob1 SE df asymp.LCL2 asymp.UCL
LVF Global 0.983 0.001 Inf 0.98 0.986
LVF Local 0.949 0.004 Inf 0.941 0.956
RVF Global 0.973 0.002 Inf 0.968 0.977
RVF Local 0.958 0.003 Inf 0.951 0.963
1 Back-transformed to probability (% correct) from logit scale
2 Confidence level: 95%


Accuracy estimates by field and level (descriptive)
field level mean_subject_percent_correct
LVF Global 97.196
LVF Local 92.338
RVF Global 95.683
RVF Local 93.52